File size: 10,864 Bytes
07001d0 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 | #!/usr/bin/env bash
# Batch test all CTA ablation variants on 4 datasets: ours / thb / ff++ / hdtf.
#
# For each ablation variant we discover its training output dir under
# outputs/cta_ablation_<VARIANT>*/checkpoints/{best.ckpt,last.ckpt}
# and run the standard test entry point with method=cta_ablation,
# overriding method.ablation_variant=<VARIANT>.
#
# Outputs:
# * One CSV per (variant, dataset) under
# outputs/cta_ablation_<VARIANT>_*/test_<dataset>_<timestamp>.csv
# (per-sample predictions; fairness CSV alongside it for the "ours" run)
# * One summary CSV listing acc/auc (and fairness scalars when available):
# cta_ablation_batch_results_<timestamp>.csv
#
# Fairness:
# * "ours" dataset (data=fairtalking) is the only one with race4 annotations,
# so it auto-emits *_fairness.csv via BaseMethod._compute_and_log_fairness_metrics.
# * thb / ff++ / hdtf only report test/acc and test/auc.
#
# Usage:
# bash scripts/batch_test_cta_ablation.sh # all discovered variants
# bash scripts/batch_test_cta_ablation.sh M1_video_only M3_intra_modal # explicit list
# bash scripts/batch_test_cta_ablation.sh --group M # only Group M
# bash scripts/batch_test_cta_ablation.sh --datasets ours thb # subset of datasets
set -eo pipefail
cd "$(dirname "$0")/.."
# load .env if present
[ -f .env ] && set -a && . ./.env && set +a
OUTPUT_DIR="outputs"
TIMESTAMP="$(date +%Y%m%d_%H%M%S)"
RESULTS_CSV="cta_ablation_batch_results_${TIMESTAMP}.csv"
# All known variants (mirror of run_all_ablations.sh)
ALL_VARIANTS=(
A1_full A2_no_asym A3_no_ltotal A4_asym_only A5_pooled_only
B1_real_only B2_all_samples B3_no_predictor
C1_no_loss_asym C2_no_loss_aux C3_no_loss_av C4_no_loss_va C5_no_detach
D1_depth_1 D2_depth_4 D3_depth_8 D4_mlp_predictor D5_shared_predictor
E1_video_freeze_0 E2_video_freeze_07 E3_video_freeze_10
F1_audio_freeze_0 F2_audio_freeze_08
G2_random_crop
M1_video_only M2_audio_only M3_intra_modal M4_noise_target M5_shuffle_pair
M6_drop_audio_infer M7_drop_video_infer
)
declare -A GROUP_VARIANTS
GROUP_VARIANTS[A]="A1_full A2_no_asym A3_no_ltotal A4_asym_only A5_pooled_only"
GROUP_VARIANTS[B]="B1_real_only B2_all_samples B3_no_predictor"
GROUP_VARIANTS[C]="C1_no_loss_asym C2_no_loss_aux C3_no_loss_av C4_no_loss_va C5_no_detach"
GROUP_VARIANTS[D]="D1_depth_1 D2_depth_4 D3_depth_8 D4_mlp_predictor D5_shared_predictor"
GROUP_VARIANTS[E]="E1_video_freeze_0 E2_video_freeze_07 E3_video_freeze_10"
GROUP_VARIANTS[F]="F1_audio_freeze_0 F2_audio_freeze_08"
GROUP_VARIANTS[G]="G2_random_crop"
GROUP_VARIANTS[M]="M1_video_only M2_audio_only M3_intra_modal M4_noise_target M5_shuffle_pair M6_drop_audio_infer M7_drop_video_infer"
# ---- args ------------------------------------------------------------------
VARIANTS=()
DATASETS=("ours" "thb" "ff++" "hdtf")
GROUP=""
DRY_RUN=false
while [[ $# -gt 0 ]]; do
case "$1" in
--group) GROUP="$2"; shift 2 ;;
--datasets)
shift
DATASETS=()
while [[ $# -gt 0 && "$1" != --* ]]; do
DATASETS+=("$1"); shift
done
;;
--dry-run) DRY_RUN=true; shift ;;
-h|--help)
grep -E '^#( |$)' "$0" | sed 's/^# \?//'
exit 0
;;
*) VARIANTS+=("$1"); shift ;;
esac
done
if [[ -n "$GROUP" ]]; then
if [[ -z "${GROUP_VARIANTS[$GROUP]+_}" ]]; then
echo "Unknown group: $GROUP. Valid: A B C D E F G M"; exit 1
fi
# Append group variants to any explicitly listed ones
for v in ${GROUP_VARIANTS[$GROUP]}; do VARIANTS+=("$v"); done
fi
if [[ ${#VARIANTS[@]} -eq 0 ]]; then
VARIANTS=("${ALL_VARIANTS[@]}")
fi
echo "============================================================"
echo "[batch_test_ablation] timestamp: $TIMESTAMP"
echo "[batch_test_ablation] results CSV: $RESULTS_CSV"
echo "[batch_test_ablation] variants: ${VARIANTS[*]}"
echo "[batch_test_ablation] datasets: ${DATASETS[*]}"
$DRY_RUN && echo "[batch_test_ablation] DRY RUN: will only show what would be tested"
echo "============================================================"
# ---- result CSV header -----------------------------------------------------
# We always include fairness columns for schema stability; non-"ours" rows
# leave them blank.
echo "variant,group,dataset,checkpoint,test_acc,test_auc,F_FPR,F_OAE,F_DP,F_MEO,run_dir,predictions_csv,timestamp" > "$RESULTS_CSV"
# ---- helpers ---------------------------------------------------------------
group_of() {
case "$1" in
A*) echo A ;;
B*) echo B ;;
C*) echo C ;;
D*) echo D ;;
E*) echo E ;;
F*) echo F ;;
G*) echo G ;;
M*) echo M ;;
*) echo "?" ;;
esac
}
# Find the most-recent training output directory for a variant.
# We accept either:
# outputs/cta_ablation_<VARIANT>/ (no timestamp suffix)
# outputs/cta_ablation_<VARIANT>_<timestamp>/ (timestamped)
find_run_dir() {
local variant="$1"
local exact="$OUTPUT_DIR/cta_ablation_${variant}"
local match
# Prefer the timestamped versions (most recent)
match=$(ls -d "$OUTPUT_DIR"/cta_ablation_${variant}_* 2>/dev/null | sort -r | head -1 || true)
if [[ -n "$match" ]]; then
echo "$match"; return
fi
if [[ -d "$exact" ]]; then echo "$exact"; return; fi
echo ""
}
# Find the best checkpoint inside a run dir; fall back to last.ckpt.
find_ckpt() {
local run_dir="$1"
local ckpt_dir="$run_dir/checkpoints"
[[ -d "$ckpt_dir" ]] || { echo ""; return; }
# Prefer best (anything not last.ckpt), pick the most-recent
local best
best=$(ls -1 "$ckpt_dir"/*.ckpt 2>/dev/null | grep -v '/last.ckpt$' | sort -r | head -1 || true)
if [[ -n "$best" ]]; then echo "$best"; return; fi
[[ -f "$ckpt_dir/last.ckpt" ]] && { echo "$ckpt_dir/last.ckpt"; return; }
echo ""
}
# Map dataset name -> hydra data config
data_cfg_for() {
case "$1" in
ours) echo "fairtalking" ;;
thb) echo "fairtalking_thb" ;;
ff++) echo "fairtalking_ffpp" ;;
hdtf) echo "fairtalking_hdtf_paired" ;;
*) echo ""; return 1 ;;
esac
}
# Extract a metric line from output. Lightning prints e.g. "test/auc 0.8932"
extract_metric() {
local key="$1" output="$2"
echo "$output" | grep -E "^[[:space:]]*${key}[[:space:]]" | tail -1 | awk '{print $NF}'
}
# Extract a fairness scalar from a fairness CSV (section=summary, metric=key).
extract_fairness_csv() {
local csv="$1" key="$2"
[[ -f "$csv" ]] || { echo ""; return; }
# Schema: section,group,n,...,metric,value
awk -F, -v k="$key" '$1=="summary" && $(NF-1)==k {print $NF; exit}' "$csv"
}
# ---- main loop -------------------------------------------------------------
for variant in "${VARIANTS[@]}"; do
grp=$(group_of "$variant")
run_dir=$(find_run_dir "$variant")
if [[ -z "$run_dir" ]]; then
echo "[$variant] NO run directory found under $OUTPUT_DIR/cta_ablation_${variant}*; skipping"
for ds in "${DATASETS[@]}"; do
echo "$variant,$grp,$ds,NO_RUN_DIR,N/A,N/A,,,,,,,$TIMESTAMP" >> "$RESULTS_CSV"
done
continue
fi
ckpt=$(find_ckpt "$run_dir")
if [[ -z "$ckpt" ]]; then
echo "[$variant] NO checkpoint in $run_dir/checkpoints; skipping"
for ds in "${DATASETS[@]}"; do
echo "$variant,$grp,$ds,NO_CKPT,N/A,N/A,,,,,$run_dir,,$TIMESTAMP" >> "$RESULTS_CSV"
done
continue
fi
echo ""
echo "============================================================"
echo "[$variant] run dir: $run_dir"
echo "[$variant] checkpoint: $ckpt"
echo "============================================================"
for ds in "${DATASETS[@]}"; do
data_cfg=$(data_cfg_for "$ds") || true
if [[ -z "$data_cfg" ]]; then
echo "[$variant] unknown dataset '$ds'; skipping"
continue
fi
ts="$(date +%Y%m%d_%H%M%S)"
pred_csv="$run_dir/test_${ds}_${ts}.csv"
echo ""
echo "[$variant] testing on dataset='$ds' (data=$data_cfg)"
echo "[$variant] predictions -> $pred_csv"
if $DRY_RUN; then
echo "[$variant][$ds] DRY RUN: skipping actual python call"
echo "$variant,$grp,$ds,$(basename "$ckpt"),DRY,DRY,,,,,$run_dir,$pred_csv,$ts" >> "$RESULTS_CSV"
continue
fi
# Run test. Hydra-style overrides: method=cta_ablation, the variant,
# data=<cfg>, the checkpoint, and force-redirect predictions CSV.
set +e
out=$(python3 src/train.py \
method=cta_ablation \
method.ablation_variant="$variant" \
data="$data_cfg" \
trainer=ddp \
backbone=timesformer \
+test_only=true \
+test_ckpt="$ckpt" \
+test_predictions_csv="$pred_csv" \
experiment_name="cta_ablation_${variant}_test_${ds}" \
2>&1)
rc=$?
set -e
# Print full output to stdout for debugging.
echo "$out"
if [[ $rc -ne 0 ]]; then
echo "[$variant][$ds] python exited with rc=$rc"
echo "$variant,$grp,$ds,$(basename "$ckpt"),ERROR,ERROR,,,,,$run_dir,$pred_csv,$ts" >> "$RESULTS_CSV"
continue
fi
acc=$(extract_metric "test/acc" "$out"); acc=${acc:-N/A}
auc=$(extract_metric "test/auc" "$out"); auc=${auc:-N/A}
# Fairness only for "ours". The fairness CSV lives next to pred_csv with
# suffix "_fairness.csv".
f_fpr=""; f_oae=""; f_dp=""; f_meo=""
if [[ "$ds" == "ours" ]]; then
fairness_csv="${pred_csv%.csv}_fairness.csv"
if [[ -f "$fairness_csv" ]]; then
f_fpr=$(extract_fairness_csv "$fairness_csv" "F_FPR")
f_oae=$(extract_fairness_csv "$fairness_csv" "F_OAE")
f_dp=$(extract_fairness_csv "$fairness_csv" "F_DP")
f_meo=$(extract_fairness_csv "$fairness_csv" "F_MEO")
echo "[$variant][$ds] fairness: F_FPR=$f_fpr F_OAE=$f_oae F_DP=$f_dp F_MEO=$f_meo"
else
echo "[$variant][$ds] fairness CSV not found at $fairness_csv (annotations missing?)"
fi
fi
echo "$variant,$grp,$ds,$(basename "$ckpt"),$acc,$auc,$f_fpr,$f_oae,$f_dp,$f_meo,$run_dir,$pred_csv,$ts" >> "$RESULTS_CSV"
echo "[$variant][$ds] acc=$acc auc=$auc"
done
done
echo ""
echo "============================================================"
echo "[batch_test_ablation] DONE. Summary CSV: $RESULTS_CSV"
echo "============================================================"
if command -v column >/dev/null 2>&1; then
column -t -s, "$RESULTS_CSV" | head -50
fi
|